Strange pattern after using "cgcurs"

[Mark.Wieringa]

Hi Tai-An,

I'm not sure if the noise you're seeing is normal for 1/3 the bandwidth, but I'll list the steps I used.

- imaged with all baselines except 1-2, full band, mfclean
- Run a phase selfcal (again excluding 1-2)
- image, excluding antenna 6 and 1-2, 1/3 band, mfclean, specify the cellsize=1.5, imsize=2048, so all images are on the same grid.

I used a robust of 0 or 0.5 and a clean cutoff of 4 times the noise in the stokes V image, which is about 6-7e-5 for 1/3 of the band.
To add the images together you use the task convol with options=final and specify the beamsize of the lowest freq image for the two higher freq images. This gets all the beam sizes the same.
After that you can add them together with maths (or linmos if you want to correct the primary beam).

Cheers,

Mark

[taianmoon]

Hi Mark,

I think I'm getting close and attached is the sub-band added map. Several things to check:

1. What is the relation between the first two (with the full band) and the last (with sub-bands) steps? Because it seems that that they are independent so basically I can do only step 3 starting from uvsplit to split the uv data into 3 sub-bands?
2. Excluding baselines/antenna means flagging their uv data using uvflag?
3. Where can I get the stokes V noise level to use in the clean cutoff?
4. In the task convol, how can I specify the beamsize of the lowest-freq image? Currently I am using the Gaussian beam fwhm and pa that were produced during restor of the lowest-freq image. However, the manual seems to say that when options=final in convol, fwhm becomes the required resolution of the output image. Should a fwhm with units of pixel number be given, rather than in units of arcsecs?

Thanks a lot again!

Tai-An

[Mark.Wieringa]

Hi Tai-An,

1. You can specify a subband using the line parameter in invert, for the first subband use: line=chan,683,1,1,1 for the second use line=chan,683,683,1,1 for the 3rd: line=chan,683,1365,1,1
This way you can use the data with selfcal applied (no need to use uvsplit)
2. No, I just use the select keyword: select=-ant(6),-ant(1)(2) to exclude antenna 6 and baseline 1-2 (shortest)
3. In invert specify stokes=i,v and give an extra output map name, you can then use sigest to get the noise in the stokes V image, this is often more accurate than the theoretical noise estimate.
4. Yes, specify the beamsize restor found for the lowest frequency and use options=final, this should make the resolution of the higher frequency images the same as the lowest one. The units are arcsec and degrees.

I think if you image the data in 3 sub-bands with the selfcal applied, use convol on the highest two frequencies and add them up you should get a result similar to what I got.

Cheers,

Mark

[taianmoon]

Hi Mark,

It turns out to be a further improvement (attached). But I still couldn't figure out that there's still some difference. Is it only the scaling is different? Or the weighting scheme is different? (I'm using robust=0 throughout all imaging; robust=0.5 is a bit noisier.) Or the selfcal clip is different (I just let it to be the default)?

This might be good enough for our scientific goals and I'll try to detect sources on it. Thanks a lot again!

Tai-An

[Mark.Wieringa]

Hi Tai-An,

I think the remaining difference might be mostly the scaling (and the fact I left out the convolve step, as I was just having a quick look).
I used the same selfcal clip (>0). There might also be some small differences in flagging. I think your image looks pretty good now, and the central area is not really limited by the strong source anymore.

Cheers,

Mark